Charlie Snell retweeted
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
783
966
10,038
2,896,953
Charlie Snell retweeted
Earlier this week, we published our technical report on Composer 2.
We're sharing additional research on how we train new checkpoints. With real-time RL, we can ship improved versions of the model every five hours.
We're sharing additional research on how we train new checkpoints. With real-time RL, we can ship improved versions of the model every five hours.
100
129
1,615
512,579
Charlie Snell retweeted
Cursor can now search millions of files and find results in milliseconds.
This dramatically speeds up how fast agents complete tasks.
We're sharing how we built Instant Grep, including the algorithms and tradeoffs behind the design.
This dramatically speeds up how fast agents complete tasks.
We're sharing how we built Instant Grep, including the algorithms and tradeoffs behind the design.
188
346
5,862
1,066,019
Charlie Snell retweeted
We're releasing a technical report describing how Composer 2 was trained.
166
477
5,027
1,310,779
Charlie Snell retweeted
Congrats to the @cursor_ai team on the launch of Composer 2!
We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support.
Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.
We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support.
Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.
509
1,353
20,035
3,573,432
Charlie Snell retweeted
It's a really good model! Excited for more people to try it
Yep, Composer 2 started from an open-source base! We will do full pretraining in the future.
Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different.
And yes, we are following the license through our inference partner terms.
22
7
221
43,397
Charlie Snell retweeted
Composer 2 marks the one-year anniversary of our large model training efforts. Since then, we've built an exceptionally talent-dense team of ~40 people with some of the best researchers and engineers from the labs, academia, industry, and more heterogeneous backgrounds.
And we are exclusively focused on coding. We don't care about models that can respond to emails, do your tax returns, or be your friend.
Every FLOP, token, parameter, and researcher is entirely dedicated to software engineering.
And we are exclusively focused on coding. We don't care about models that can respond to emails, do your tax returns, or be your friend.
Every FLOP, token, parameter, and researcher is entirely dedicated to software engineering.
Composer 2 is now available in Cursor.
74
52
1,021
139,624
Charlie Snell retweeted
now that i can talk about it, i had early access to composer 2 and have been using it for the past few days. quick thoughts:
- feels very close to gpt-5.4-high at lesser price
- follows instructions really well
- holds structure over long threads without much context rot
- very noticeable leap from composer-1.5 but also codex 5.3 and opus 4.6 low/medium in reasoning + code quality
- "fast" genuinely fast (2-3x faster than the non-fast variant), beats opus-4.6-high on code throughput
cursor cooked. absolutely banger launch that's gonna save people serious money
- feels very close to gpt-5.4-high at lesser price
- follows instructions really well
- holds structure over long threads without much context rot
- very noticeable leap from composer-1.5 but also codex 5.3 and opus 4.6 low/medium in reasoning + code quality
- "fast" genuinely fast (2-3x faster than the non-fast variant), beats opus-4.6-high on code throughput
cursor cooked. absolutely banger launch that's gonna save people serious money
Composer 2 is now available in Cursor.
9
8
190
15,712
Charlie Snell retweeted
We trained Composer to self-summarize through RL instead of a prompt.
This reduces the error from compaction by 50% and allows Composer to succeed on challenging coding tasks requiring hundreds of actions.
This reduces the error from compaction by 50% and allows Composer to succeed on challenging coding tasks requiring hundreds of actions.
90
98
1,642
230,245
Charlie Snell retweeted
We're sharing a new method for scoring models on agentic coding tasks.
Here's how models in Cursor compare on intelligence and efficiency:
Here's how models in Cursor compare on intelligence and efficiency:
202
252
2,853
637,188
Charlie Snell retweeted
See you Sunday!
Participants will get free Cursor credits and the winners will get a year of Cursor Ultra.
Cursor team will be on hand.
Participants will get free Cursor credits and the winners will get a year of Cursor Ultra.
Cursor team will be on hand.
Our hackathon on Sunday is gonna be HUGE
We have many participants from every major AI lab, and sonmany sick speakers like @cHHillee
Power to Prefill, Dirt to Decode, A Full-Stack AI Infrastructure Hackathon
Limited spots, apply now
luma.com/SAxFSHack
1
3
42
23,448
Charlie Snell retweeted
We believe Cursor discovered a novel solution to Problem Six of the First Proof challenge, a set of math research problems that approximate the work of Stanford, MIT, Berkeley academics. Cursor's solution yields stronger results than the official, human-written solution.
Notably, we used the same harness that built a browser from scratch a few weeks ago. It ran fully autonomously, without nudging or hints, for four days.
This suggests that our technique for scaling agent coordination might generalize beyond coding.
Notably, we used the same harness that built a browser from scratch a few weeks ago. It ran fully autonomously, without nudging or hints, for four days.
This suggests that our technique for scaling agent coordination might generalize beyond coding.
258
495
8,186
1,054,918
Charlie Snell retweeted
Cursor now shows you demos, not diffs.
Agents can use the software they build and send you videos of their work.
Agents can use the software they build and send you videos of their work.
392
584
8,289
4,144,412
